A method for intelligent power emergency load shedding based on optimized neural network parameters

Abstract

The present invention provides an emergency intelligent load shedding method for electric power based on optimizing neural network parameters, which relates to the technical field of power grids, including: establishing a data set of a power system bus based on reciprocal voltage sensitivity and frequency variation; obtaining an improved dung beetle optimization algorithm and an improved artificial neural network model based on an improved Tent chaotic mapping and adaptive t-distribution perturbation, and training the improved artificial neural network model based on the data set to obtain a trained artificial neural network model; identifying the load shedding state and non-load shedding state of each power system bus based on the trained artificial neural network model, and obtaining a load shedding strategy based on the load shedding state and non-load shedding state. The present invention improves the traditional dung beetle optimization algorithm, and applies the improved algorithm to the traditional artificial neural network model to obtain an artificial neural network model with higher recognition accuracy, thereby improving the efficiency of the power grid load shedding strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of power grids, and in particular to an emergency intelligent load shedding method for electric power based on optimizing neural network parameters. Background Art

[0002] As the scale of power grids expands, power systems become more and more complex, and it is critical to verify whether the power system remains stable and whether its parameters are set within acceptable thresholds. At present, there have been many power system instability events, and load shedding operations are required to maintain stable operation of the power system. However, the traditional load shedding method has poor system accuracy when responding to emergencies. Therefore, it is of great significance to study a novel and effective emergency load shedding method.

[0003] Common existing emergency load shedding methods for power supply include low-frequency load shedding and low-voltage load shedding. Although low-voltage load shedding is a popular and cost-effective method for stabilizing voltage, it is difficult to determine the location, timing, and calculation of load shedding values. Traditional low-frequency load shedding strategies will cut a fixed number of loads according to each frequency setting, which is greatly affected by frequency parameters. In addition, solving the load shedding problem through intelligent optimization algorithms is also an important idea. Currently, the commonly used optimization algorithm is mainly the particle swarm optimization algorithm, but it has problems such as local optimality and slow optimization speed, resulting in traditional low-frequency load shedding, low-voltage load shedding, and intelligent optimization algorithm methods cannot respond to emergency load shedding needs in a timely manner, and the load shedding strategy is inefficient. Therefore, it is urgent to provide a solution to improve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide an emergency intelligent load shedding method for electric power based on optimizing neural network parameters, so as to improve the problems of the existing low-voltage load shedding method, such as the difficulty in calculating the load shedding value, the low-frequency load shedding method being greatly affected by the frequency, the local optimum and slow optimization speed of the intelligent optimization algorithm resulting in low efficiency of the load shedding strategy.

[0005] The present invention provides an electric power emergency intelligent load shedding method based on optimizing neural network parameters, which adopts the following technical solutions:

[0006] Establishing a data set of the power system bus based on the reciprocal voltage sensitivity and the frequency variation, wherein the data set consists of load power, generator power, transmission power, reciprocal voltage, frequency, load shedding distribution, voltage sensitivity and load shedding index of the power system bus;

[0007] Based on the improved Tent chaotic mapping and the adaptive t-distribution perturbation strategy, an improved chaotic dung beetle optimization algorithm is obtained, and the number of neurons in the initial artificial neural network model is optimized based on the improved chaotic dung beetle optimization algorithm to obtain an improved artificial neural network model, and the improved artificial neural network model is trained based on the data set to obtain a trained artificial neural network model;

[0008] The load shedding state and the non-load shedding state of each power system bus are identified based on the trained artificial neural network model, and a load shedding strategy is acquired based on the load shedding state and the non-load shedding state.

[0009] The beneficial effects of the power emergency intelligent load shedding method based on optimizing neural network parameters provided by the present invention are as follows: firstly, the traditional dung beetle optimization algorithm is improved based on the improved Tent chaotic mapping and the adaptive t-distribution perturbation strategy, which can improve the problem that when the traditional Tent chaotic mapping initializes the sequence population, the Tent chaotic sequence is easily caused to fall into small period points and unstable period points during iteration, so that the algorithm can converge quickly and avoid falling into local optimality; secondly, the algorithm's global search and local optimal jump capabilities can be enhanced through the adaptive t-distribution mutation perturbation strategy; finally, the weights of the improved artificial neural network are trained using a data set, so that an artificial neural network model with better performance can be obtained, thereby improving the accuracy of using the model to identify the load shedding and non-load shedding states of the power system bus, and ultimately improving the efficiency of the load shedding strategy.

[0010] Optionally, the process of obtaining the load shedding distribution amount includes:

[0011] Obtaining minimum load power based on frequency variation, active power imbalance and secondary control power;

[0012] Obtaining a load shedding order based on a first-order derivative of the charge quantity with respect to the mutual-sensitivity voltage, and obtaining a relative voltage sensitivity of a preselected bus based on the order of arrangement;

[0013] A load shedding index is obtained based on the relative voltage sensitivity, and a load shedding distribution amount is obtained based on the load shedding index and the minimum load power.

[0014] Optionally, the process of obtaining an improved chaotic dung beetle optimization algorithm based on the improved Tent chaotic mapping and the adaptive t-distribution perturbation strategy includes:

[0015] Step 1: Divide the dung beetle population into four subgroups of rolling, breeding, foraging and stealing based on the population distribution ratio;

[0016] Step 2: Initializing the maximum number of iterations, and initializing the position of each dung beetle in the dung beetle sub-swarm based on the improved Tent chaotic map, and then obtaining the position of the first dung beetle population;

[0017] Step 3: Generate a random number and preset a probability operator. If the random number is less than the probability operator, the position of the first dung beetle population is iteratively updated based on the adaptive t-distribution perturbation strategy and the greedy rule to obtain the position of the second dung beetle population; if the random number is greater than or equal to the probability operator, the position of the first dung beetle population is iteratively updated based on the position update strategy to obtain the position of the third dung beetle population, wherein the position of the third dung beetle population consists of the updated rolling dung beetle position, the updated breeding dung beetle position, the updated foraging dung beetle position, and the updated stealing dung beetle position;

[0018] Step 4: Determine whether the current number of iterations reaches the maximum number of iterations. If so, output the optimal fitness function value and the optimal dung beetle position. Otherwise, return to step 3 and continue to execute steps 3 to 4 until the maximum number of iterations is reached and the iteration is stopped. The optimal dung beetle position consists of the second dung beetle population position and the third dung beetle population position.

[0019] Optionally, the mathematical expression of the improved Tent chaos map is: ;

[0020] in, Represents a random number between 0 and 1. represents the number of dung beetle populations, represents the location of the first dung beetle population, represents the position of the dung beetle population at the initial moment, represents a random variable.

[0021] Optionally, the process of obtaining the second dung beetle population position after iteratively updating the first dung beetle population position based on the adaptive t-distribution perturbation strategy and the greedy rule includes:

[0022] After iteratively updating the position of the dung beetle population at the initial moment based on the adaptive t-distribution perturbation strategy, the perturbed position of the dung beetle population is obtained;

[0023] Based on the greedy rule, the fitness function value of the first dung beetle population position and the fitness function value of the disturbed dung beetle population position are compared, and the dung beetle population position corresponding to the smaller fitness function value is selected as the second dung beetle population position, wherein the second dung beetle population position is one of the first dung beetle population position and the disturbed dung beetle population position.

[0024] Optionally, the mathematical expression for the position of the dung beetle population after disturbance is: ;

[0025] in, Indicates the current iteration number, The dung beetle population after the disturbance is The position at the iteration, represents the number of iterations, represents a t-distribution with the number of algorithm iterations as parameter’s degrees of freedom.

[0026] Optionally, training the improved artificial neural network model based on the data set includes:

[0027] Initializing the weights of the improved artificial neural network, obtaining the initial weights of neurons in the hidden layer and the output layer, and inputting the data set into the improved artificial neural network model to obtain a predicted output value;

[0028] A loss function is constructed based on the difference between the predicted output value and the actual output value, and the optimal weights of the hidden layer and output layer neurons corresponding to the minimum or little change of the loss function are respectively obtained based on the gradient descent method.

[0029] Optionally, the process of training the improved artificial neural network model based on the data set is actually a process of training the initial weights of the neurons inside the improved artificial neural network model to obtain the optimal weights, wherein the neurons are composed of hidden layer neurons and output layer neurons, the initial weights are composed of the initial weights of the hidden layer neurons and the initial weights of the output layer neurons, and the optimal weights are composed of the optimal weights of the hidden layer neurons and the optimal weights of the output layer neurons.

[0030] Optionally, the loss function is equivalent to a mean square error function or a fitness function value, and the mathematical expression is:

[0031] ;

[0032] in, represents the fitness function value, represents the loss function value, represents the mean square error function, Represents the number of samples in the dataset, is the desired output target, is the prediction result of the trained artificial neural network model. and is the activation function of the neuron, and are the weights of neurons in the hidden and output layers, is the data set input to the trained artificial neural network model.

[0033] Optionally, the process of acquiring a load shedding strategy based on the load shedding state and the non-load shedding state includes:

[0034] Acquire a target power system bus in a load shedding state based on the power system bus;

[0035] The load of the target power system bus is cut off until the target power system bus is in a non-load-cut state. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of a method for power emergency intelligent load shedding based on optimizing neural network parameters provided by the present invention;

[0037] Figure 2 It is a flow chart of iterative optimization of the improved chaotic dung beetle optimization algorithm provided by the present invention;

[0038] Figure 3 It is a structural diagram of the improved artificial neural network model provided by the present invention combined with the improved chaotic dung beetle optimization algorithm;

[0039] Figure 4 It is a flow chart of load shedding training data collection and load shedding strategy generation provided by the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0041] The embodiment of the present invention provides a method for intelligent emergency load shedding based on optimizing neural network parameters. Figure 1 , including the following steps:

[0042] S1. Establishing a data set of a power system bus based on reciprocal voltage sensitivity and frequency variation, wherein the data set consists of load power, generator power, transmission power, reciprocal voltage, frequency, load shedding distribution, voltage sensitivity and load shedding index of the power system bus;

[0043] S2. Obtaining an improved chaotic dung beetle optimization algorithm based on an improved Tent chaotic mapping and an adaptive t-distribution perturbation strategy, optimizing the number of neurons in the initial artificial neural network model based on the improved chaotic dung beetle optimization algorithm, obtaining an improved artificial neural network model, and training the improved artificial neural network model based on the data set to obtain a trained artificial neural network model;

[0044] S3. Identify the load shedding state and the non-load shedding state of each power system bus based on the trained artificial neural network model, and acquire a load shedding strategy based on the load shedding state and the non-load shedding state.

[0045] In some embodiments, when executing step S1 to obtain a data set of the power system bus, the data set is load shedding data, and the data set is actually composed of 328 training sample sets including load shedding data and non-load data, which are tested on the IEEE-37 bus sample at a load level of 60% to 100%, and are used for training the improved artificial neural network model in step S2.

[0046] In some embodiments, see Figure 4 , which represents a flow chart of load shedding training data collection and load shedding strategy generation provided by the present invention, specifically: testing is performed on a standard IEEE-37 bus sample with a load level of 60%, and it is determined whether the test load level reaches 100%. If the test load level does not reach 100%, the power system completely collapses and the generator will be out of power. After measuring and collecting system parameters, it is determined whether the frequency is less than 59.7HZ. If the frequency is less than 59.7HZ, a load shedding strategy is constructed for each load level, otherwise the load level is increased until the load level reaches 100%.

[0047] In some embodiments, the process of obtaining the load shedding distribution amount in step S1 includes:

[0048] S1-1, obtaining the minimum load power based on the frequency change, the active power imbalance and the secondary control power;

[0049] S1-2, obtaining the order of load shedding based on the first-order derivative of the charge amount with respect to the mutual-sensitivity voltage, and obtaining the relative voltage sensitivity of the preselected bus based on the order of arrangement;

[0050] S1-3. Obtain a load shedding index based on the relative voltage sensitivity, and obtain a load shedding distribution amount based on the load shedding index and the minimum load power.

[0051] Specifically, when executing step S1-1, the mathematical expression of the minimum load power is: ;

[0052] in, is the allowable frequency variation, is the active power unbalance, About The weight coefficient of is the power used for secondary control, Indicates the minimum load power.

[0053] In fact, if there is a power shortage or imbalance between the power supply and the load, a frequency deviation will occur. In order to correct this situation, the frequency control is performed in a specific order: primary control first, then secondary control. If the frequency has not returned to the allowed level after the secondary control, the load is removed.

[0054] Furthermore, when executing step S1-2, the first-order derivative of the charge amount to the mutual sensitivity voltage is expressed as a reciprocal voltage sensitivity equation, specifically: ;

[0055] in, To enter The amount of charge, , For Node and The voltage at For Node The self-admittance, For Node and nodes The mutual admittance between for and The phase angle difference of the voltage at For segment The phase angle of the total conductance.

[0056] In fact, when executing step S1-2 to obtain the order of load shedding, Determine the order of load shedding, The bus with the largest value is listed at the top. The bus with the smallest value is listed at the bottom, in descending order.

[0057] Further, when executing step S1-2 to obtain the relative voltage sensitivity of the pre-selected bus, the relative voltage sensitivity is expressed as: ;

[0058] in, Relative voltage sensitivity, is the number of power system buses.

[0059] Specifically, when executing step S1-3 to obtain the load shedding index based on the relative voltage sensitivity, the load shedding index is expressed as: ;

[0060] in, Indicates load shedding index.

[0061] Further, when executing step S1-3 to obtain the load shedding distribution amount based on the load shedding index and the minimum load power, the load shedding distribution amount is expressed as: ;

[0062] in, Indicates the load shedding distribution amount.

[0063] In some embodiments, see Figure 2 , represents a flowchart of iterative optimization of the improved chaotic dung beetle optimization algorithm provided by the present invention, and the specific process is: initialize the maximum number of iterations and the number of dung beetle populations, use the improved Tent chaos to map the position of each dung beetle in the dung beetle population and calculate the fitness function value of the dung beetle population, update the dung beetle position based on the fitness function value and judge whether the dung beetle position is out of bounds, if not out of bounds, continue to judge whether the random number is less than the preset probability operator, if less than, perform adaptive t distribution mutation to update the dung beetle position to obtain the disturbed dung beetle population position, and use the greedy rule to filter the disturbed dung beetle population position, obtain the optimal solution (optimal dung beetle population position) and the optimal fitness value, and output the optimal dung beetle position as the result after the maximum number of iterations is reached.

[0064] In some embodiments, when executing step S2, the process of obtaining an improved chaotic dung beetle optimization algorithm based on the improved Tent chaotic mapping and the adaptive t-distribution perturbation strategy includes:

[0065] S2-1, based on the population distribution ratio, the dung beetle population is divided into four subgroups: rolling, breeding, foraging and stealing;

[0066] S2-2, initializing the maximum number of iterations, and initializing the position of each dung beetle in the dung beetle sub-swarm based on the improved Tent chaotic map, and then obtaining the position of the first dung beetle population;

[0067] S2-3, generate a random number and preset a probability operator, if the random number is less than the probability operator, iteratively update the position of the first dung beetle population based on the adaptive t-distribution perturbation strategy and the greedy rule to obtain the position of the second dung beetle population; if the random number is greater than or equal to the probability operator, iteratively update the position of the first dung beetle population based on the position update strategy to obtain the position of the third dung beetle population, the third dung beetle population position is composed of the updated rolling dung beetle position, the updated breeding dung beetle position, the updated foraging dung beetle position, and the updated stealing dung beetle position;

[0068] S2-4, determine whether the current number of iterations reaches the maximum number of iterations. If so, output the optimal fitness function value and the optimal dung beetle position. Otherwise, return to step 3 and continue to execute steps 3 to 4 until the maximum number of iterations is reached and the iteration is stopped, wherein the optimal dung beetle position is composed of the second dung beetle population position and the third dung beetle population position.

[0069] Specifically, when executing step S2-2, the improved Tent chaotic mapping introduces a random variable on the basis of the traditional Tent chaotic mapping. The mathematical expression of the improved Tent chaos map is obtained as follows:

[0070] ;

[0071] in, Represents a random number between 0 and 1. represents the number of dung beetle populations, represents the location of the first dung beetle population, represents the position of the dung beetle population at the initial moment, represents a random variable.

[0072] In fact, using the Tent chaotic map to initialize the position of the dung beetle population and using the ergodicity of the Tent to generate a more evenly distributed chaotic sequence can reduce the impact of the initial value on the algorithm optimization. The traditional Tent chaotic map is prone to fall into the small periodic points and unstable periodic points of the chaotic sequence during iteration. By introducing random variables , which can improve the above problems by increasing the uniformity of the initial dung beetle population position traversal while maintaining the randomness, ergodicity and regularity of the chaotic mapping.

[0073] Specifically, when executing step S2-3, the mathematical expressions of the position of the third dung beetle population are as follows:

[0074] The updated position of the rolling dung beetle is expressed as: ;

[0075] in, Indicates the current iteration number of the rolling ball dung beetle, Indicates The ball-rolling dung beetle The position at the iteration, Indicates The ball-rolling dung beetle The position at the iteration, Indicates The ball-rolling dung beetle The position at the iteration, represents the disturbance coefficient, For the interval Random values, is the natural coefficient, assigned a value of 1, represents the global worst position, is the factor that simulates the change of light intensity.

[0076] The updated breeding dung beetle position is represented as: ;

[0077] in, Indicates the current iteration number of breeding dung beetles, represents the best position for breeding dung beetles in the current iteration, No. The first iteration The location of the ovum, Indicates The first iteration The location of the ovum, and represents two independent random vectors of size 1 × D, where D represents the dimension, and They represent the optimal lower bound and upper bound of the breeding dung beetle location respectively.

[0078] The updated foraging dung beetle position is represented by: ;

[0079] in, represents the current iteration number of the foraging dung beetle, , , represents the global best feeding position, and They represent the lower and upper bounds of the optimal feeding area of ​​the foraging dung beetle, and denote the lower and upper bounds of the optimization problem, respectively. Indicates The first iteration Only foraging for the location of dung beetles, Indicates The first iteration Only the location of the dung beetle is foraging, represents a random number that follows a normal distribution, represents a random vector of (0,1), Indicates the maximum number of iterations.

[0080] The updated position of the stealing dung beetle is represented as: ;

[0081] in, Represents the current iteration number of the stealing dung beetle, represents the best position of the stealing dung beetle in the current iteration, Indicates A thieving dung beetle The position at the iteration, Indicates A thieving dung beetle The position at the iteration, represents a random vector of size 1×D, which obeys the normal distribution, Indicates a constant value.

[0082] In fact, when executing step S2-3, the value of the random number is between 0 and 1, and the value of the probability operator is preset.

[0083] Furthermore, when executing step S2-3, the process of iteratively updating the first dung beetle population position based on the adaptive t-distribution perturbation strategy and the greedy rule to obtain the second dung beetle population position includes:

[0084] S2-3-1, after iteratively updating the position of the dung beetle population at the initial moment based on the adaptive t-distribution perturbation strategy, the position of the dung beetle population after the perturbation is obtained;

[0085] S2-3-2. Compare the fitness function value of the first dung beetle population position and the fitness function value of the disturbed dung beetle population position based on the greedy rule, and select the dung beetle population position corresponding to the smaller fitness function value as the second dung beetle population position, wherein the second dung beetle population position is one of the first dung beetle population position and the disturbed dung beetle population position.

[0086] Specifically, when executing step S2-3-1, the mathematical expression of the position of the dung beetle population after disturbance is:

[0087] ;

[0088] in, Indicates the current iteration number, The dung beetle population after the disturbance is The position at the iteration, Indicates that the first dung beetle population is in The position of the iteration, represents the number of iterations, represents a t-distribution with the number of algorithm iterations as parameter’s degrees of freedom.

[0089] Furthermore, when executing step S2-3-2, the mathematical expression of the greedy rule is: ;

[0090] in, is the position of the dung beetle population after the greedy rule update, represents the fitness function value of the dung beetle population position after the disturbance, Represents the fitness function value of the first dung beetle population position.

[0091] In some embodiments, performing step S2 to train the improved artificial neural network model based on the data set includes:

[0092] S2-5, initializing the weights of the improved artificial neural network, obtaining the initial weights of neurons in the hidden layer and the output layer, and inputting the data set into the improved artificial neural network model to obtain a predicted output value;

[0093] S2-6. Construct a loss function based on the difference between the predicted output value and the actual output value, and obtain the optimal weights of the hidden layer and output layer neurons corresponding to the minimum or little change of the loss function based on the gradient descent method.

[0094] In some embodiments, the process of executing step S2 to train the improved artificial neural network model based on the data set is actually a process of training the initial weights of the neurons inside the improved artificial neural network model to obtain the optimal weights, wherein the neurons are composed of hidden layer neurons and output layer neurons, and the initial weights are composed of the initial weights of the hidden layer neurons. and the initial weights of the output layer neurons The optimal weight is composed of the optimal weight of the hidden layer neurons and the optimal weights of the output layer neurons composition.

[0095] In some embodiments, see Figure 3 , which represents the structure diagram of the improved artificial neural network model combined with the improved chaotic dung beetle optimization algorithm provided by the present invention, specifically: the improved artificial neural network model consists of an input layer, a hidden layer and an output layer, and five power parameters: the voltage on each system bus ,frequency , the transmission power of each system bus , generator power , load power The power parameters , input these five power parameters into the hidden layer for feature extraction, and obtain the optimal weights of the output layer neurons , and the initial weights of the hidden layer neurons are adjusted based on the improved dung beetle optimization algorithm Train to obtain the optimal hidden layer neuron weights Finally, after passing through an output layer, we get an output variable , output variable Indicates load shedding or non-load shedding status.

[0096] The load shedding state means that among all system buses, part of the load needs to be cut off to maintain the bus state of the entire power system, which is recorded as 1; the non-load shedding state means that among all system buses, part of the load does not need to be cut off to maintain the bus state of the entire power system, which is recorded as 0.

[0097] Specifically, when executing step S2-5, the initial weights of the hidden neurons , the initial weights of the output layer neurons , the optimal weights of the hidden layer neurons , the optimal weights of the output layer neurons The activation functions are The activation function is, therefore,

[0098] All must be limited to certain ranges, including: ; ;in, is the number of connections between neurons.

[0099] Furthermore, in executing step S2-6, the process of obtaining the optimal weights of the hidden layer and output layer neurons corresponding to the time when the loss function is minimized or does not change much based on the gradient descent method includes: in each training process, using the gradient descent method to derive the loss function to obtain the weights of the hidden layer and output layer neurons in each training, until the loss function value is minimized or does not change much in a certain training, and the training is stopped, and the optimal weights of the hidden layer neurons are output. and the optimal weights of the output layer neurons .

[0100] Among them, the loss function is equivalent to the fitness function value or the mean square error function , the three have the same mathematical expression.

[0101] In some embodiments, when executing step S2-6, the loss function is equivalent to the mean square error function or the fitness function value, and the mathematical expression is: ;

[0102] in, represents the fitness function value, represents the loss function value, represents the mean square error function, Represents the number of samples in the dataset, is the desired output target, is the prediction result of the trained artificial neural network model. and is the activation function of the neuron, and are the weights of neurons in the hidden and output layers, is the data set input to the trained artificial neural network model.

[0103] In some embodiments, when executing step S3, the process of acquiring the load shedding strategy based on the load shedding state and the non-load shedding state includes:

[0104] Acquire a target power system bus in a load shedding state based on the power system bus;

[0105] The load of the target power system bus is cut off until the target power system bus is in a non-load-cut state.

[0106] Specifically, the number of the target power system buses is one or more.

[0107] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.

Claims

1. A method for intelligent power emergency load shedding based on optimization of neural network parameters, characterized in that: include: S001: Establishing a data set of a power system bus based on reciprocal voltage sensitivity and frequency variation, wherein the data set consists of load power, generator power, transmission power, reciprocal voltage, frequency, load shedding distribution, voltage sensitivity and load shedding index of the power system bus; S002: Obtain the minimum load power based on the frequency change, the active power imbalance and the secondary control power. The mathematical expression of the minimum load power is: ； in, is the allowable frequency variation, is the active power unbalance, About The weight coefficient of is the power used for secondary control, Indicates the minimum load power; S003: Obtaining the order of load shedding based on the first-order derivative of the charge quantity to the mutual-sensitivity voltage, and obtaining the relative voltage sensitivity of the preselected bus based on the order of arrangement, wherein: When obtaining the order of load shedding, use Determine the order of load shedding, The bus with the largest value is listed at the top. The bus with the smallest value is listed at the bottom, in descending order; The first-order derivative of the charge to the mutual sensitivity voltage is expressed as a reciprocal voltage sensitivity equation, specifically: ； in, To enter The amount of charge, , For Node and The voltage at For Node The self-admittance, For Node and nodes The mutual admittance between for and The phase angle difference of the voltage at For segment The phase angle of the total conductance; The relative voltage sensitivity is expressed as: ; in, Relative voltage sensitivity, is the number of power system buses; S004: Obtaining a load shedding index based on the relative voltage sensitivity, and obtaining a load shedding distribution amount based on the load shedding index and the minimum load power, wherein: The load shedding index is expressed as: ; in, Indicates load shedding index; The load shedding distribution amount is expressed as: ; in, Indicates the load shedding distribution amount; S005: Divide the dung beetle population into four dung beetle subgroups of rolling, breeding, foraging and stealing based on the population distribution ratio, initialize the maximum number of iterations, and initialize the position of each dung beetle in the dung beetle subgroup based on the improved Tent chaotic map, and then obtain the first dung beetle population position, wherein the mathematical expression of the improved Tent chaotic map is: ; in, Represents a random number between 0 and 1. represents the number of dung beetle populations, represents the location of the first dung beetle population, represents the position of the dung beetle population at the initial moment, represents a random variable; S006: Generate a random number and preset a probability operator. If the random number is less than the probability operator, the position of the first dung beetle population is iteratively updated based on the adaptive t-distribution perturbation strategy and the greedy rule to obtain the position of the second dung beetle population; if the random number is greater than or equal to the probability operator, the position of the first dung beetle population is iteratively updated based on the position update strategy to obtain the position of the third dung beetle population, wherein the position of the third dung beetle population is composed of the updated rolling dung beetle position, the updated breeding dung beetle position, the updated foraging dung beetle position, and the updated stealing dung beetle position; wherein the mathematical expression of the disturbed dung beetle population position is: ； in, Indicates the current iteration number, The dung beetle population after the disturbance is The position at the iteration, represents the number of iterations, represents the t distribution with the number of algorithm iterations as the parameter degree of freedom; S007: Determine whether the current number of iterations reaches the maximum number of iterations. If so, output the optimal fitness function value and the optimal dung beetle position. Otherwise, continue iterating until the maximum number of iterations is reached and then stop iterating. The optimal dung beetle position is composed of the second dung beetle population position and the third dung beetle population position. S008: Optimizing the number of neurons in the initial artificial neural network model to obtain an improved artificial neural network model, and training the improved artificial neural network model based on the data set to obtain a trained artificial neural network model; S009: Based on the trained artificial neural network model, the load shedding state and non-load shedding state of each power system bus are identified, and the load shedding strategy is obtained based on the load shedding state and non-load shedding state. The load shedding strategy includes: testing on a standard IEEE-37 bus sample with a load level of 60% to determine whether the test load level reaches 100%. If the test load level does not reach 100%, the power system will completely collapse and the generator will be out of power. After measuring and collecting system parameters, it is determined whether the frequency is less than 59.7HZ. If the frequency is greater than or equal to 59.7HZ, the load level is increased until the test load reaches 100%. If the frequency is less than 59.7HZ, the load level is increased based on the load shedding distribution until the test load is less than 100%.

2. The method for emergency intelligent load shedding based on optimizing neural network parameters according to claim 1 is characterized in that: The process of iteratively updating the first dung beetle population position based on the adaptive t-distribution perturbation strategy and the greedy rule to obtain the second dung beetle population position includes: After iteratively updating the position of the dung beetle population at the initial moment based on the adaptive t-distribution perturbation strategy, the perturbed position of the dung beetle population is obtained; Based on the greedy rule, the fitness function value of the first dung beetle population position and the fitness function value of the disturbed dung beetle population position are compared, and the dung beetle population position corresponding to the smaller fitness function value is selected as the second dung beetle population position, wherein the second dung beetle population position is one of the first dung beetle population position and the disturbed dung beetle population position.

3. The method for emergency intelligent load shedding based on optimizing neural network parameters according to claim 1 is characterized in that: Training the improved artificial neural network model based on the data set includes: Initializing the weights of the improved artificial neural network, obtaining the initial weights of neurons in the hidden layer and the output layer, and inputting the data set into the improved artificial neural network model to obtain a predicted output value; A loss function is constructed based on the difference between the predicted output value and the actual output value, and the optimal weights of the hidden layer and output layer neurons corresponding to the minimum or little change of the loss function are respectively obtained based on the gradient descent method.

4. The method for emergency intelligent load shedding based on optimizing neural network parameters according to claim 1 is characterized in that: The process of training the improved artificial neural network model based on the data set is actually a process of training the initial weights of the neurons inside the improved artificial neural network model to obtain the optimal weights, wherein the neurons are composed of hidden layer neurons and output layer neurons, the initial weights are composed of the initial weights of the hidden layer neurons and the initial weights of the output layer neurons, and the optimal weights are composed of the optimal weights of the hidden layer neurons and the optimal weights of the output layer neurons.

5. The method for emergency intelligent load shedding based on optimizing neural network parameters according to claim 3 is characterized in that: The loss function is equivalent to the mean square error function or the fitness function value, and the mathematical expression is: ； in, represents the fitness function value, represents the loss function value, represents the mean square error function, m represents the number of samples in the dataset, is the desired output target, is the prediction result of the trained artificial neural network model. and is the activation function of the neuron, and are the weights of neurons in the hidden and output layers, is the data set input to the trained artificial neural network model.

Images